Module hybrid 

Hybrid search: RRF fusion of semantic + BM25, then boosts and rerank.

Port of ~/src/semble/src/semble/search.py. Three entry points:

• search_semantic — cosine similarity over the dense index.
• search_bm25 — BM25 scoring (re-exported from the bm25 module).
• search_hybrid — fuses both ranked lists via Reciprocal Rank Fusion (k=60), over-fetching top_k * 5 candidates, then applies ripvec’s boost_multi_chunk_files + apply_query_boost + the penalty-aware rerank_topk.

Constants

RRF_K

Reciprocal Rank Fusion smoothing constant. Matches Python _RRF_K = 60 from search.py:11.

Functions

parallel_sgemv

Parallel matrix-vector multiply via row-sharded BLAS sgemv.

search_hybrid

Hybrid search: alpha-weighted RRF fusion of semantic + BM25, followed by file-coherence + query boosts and the penalty-aware reranker. Mirrors search.py:search_hybrid.

search_semantic

Pure semantic search: rank every chunk by dot product against the query embedding, then take the top-k after optional selector mask.